Abstract
Accurate mapping of next-generation sequencing (NGS) reads to reference genomes is crucial for almost all NGS applications and downstream analyses. Various repetitive elements in human and other higher eukaryotic genomes contribute in large part to ambiguously (non-uniquely) mapped reads. Most available NGS aligners attempt to address this by either removing all non-uniquely mapping reads, or reporting one random or "best" hit based on simple heuristics. Accurate estimation of the mapping quality of NGS reads is therefore critical albeit completely lacking at present. Here we developed a generalized software toolkit "AlignerBoost", which utilizes a Bayesian-based framework to accurately estimate mapping quality of ambiguously mapped NGS reads. We tested AlignerBoost with both simulated and real DNA-seq and RNA-seq datasets at various thresholds. In most cases, but especially for reads falling within repetitive regions, AlignerBoost dramatically increases the mapping precision of modern NGS aligners without significantly compromising the sensitivity even without mapping quality filters. When using higher mapping quality cutoffs, AlignerBoost achieves a much lower false mapping rate while exhibiting comparable or higher sensitivity compared to the aligner default modes, therefore significantly boosting the detection power of NGS aligners even using extreme thresholds. AlignerBoost is also SNP-aware, and higher quality alignments can be achieved if provided with known SNPs. AlignerBoost’s algorithm is computationally efficient, and can process one million alignments within 30 seconds on a typical desktop computer. AlignerBoost is implemented as a uniform Java application and is freely available at https://github.com/Grice-Lab/AlignerBoost.
Highlights
Numerous genome-scale experimental applications are possible due to the advent of high throughput, low cost next-generation sequencing (NGS) platforms, including genome sequencing/re-sequencing, gene expression profiling, mRNA splicing prediction/characterization, single nucleotide polymorphism (SNP) identification and genotyping, and disease-associated variant identification
To calculate the mapping precision and sensitivity, a “correct-mapping” is defined as aligned boundaries that are within +/- 20% of the true locus relative to the alignment length
Though ultra-fast speed has been achieved in many state-of-art NGS aligners, rarely have there been attempts to improve the mapping quality in terms of precision and sensitivity
Summary
Numerous genome-scale experimental applications are possible due to the advent of high throughput, low cost next-generation sequencing (NGS) platforms, including genome sequencing/re-sequencing, gene expression profiling, mRNA splicing prediction/characterization, SNP identification and genotyping, and disease-associated variant identification. The most commonly used algorithms for seed-search are Hash-index (e.g. MAQ, GSNAP, SRMapper, mrsFAST-Ultra, SeqAlto [1,2,3,4,5]), "Burrows-Wheeler Transform" (e.g. Bowtie/Bowtie, BWA, SOAP2 [6,7,8,9,10]), un-compressed tries (e.g. STAR [11]), or a mixture of the above (e.g. YOABS [12]) These seed-search algorithms usually use relatively small segments of the reads ("seeds") to initiate mapping, due to large RAM requirements to build the index. Some human pseudogene classes may have more than 500 copies of over 3,000 bp; a few human SINE retrotransposon families may have over 100,000 copies of about 300 bp In these repetitive regions, a “low complexity” seed might not even exist, leading to biased mapping in favor of these regions and subsequent false mapping
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call